By 2026 most cover letters in a recruiter inbox are at least partly AI-drafted. ChatGPT with a job description pasted in and a one-line prompt produces a predictable shape: a formula opener, generic enthusiasm in the body, no specifics about the company, and a templated three-paragraph structure that closes with the same sign-off across thousands of applicants. This guide is the screening playbook for hiring managers and recruiters. Five cover-letter-specific tells, a five-step verification HowTo, the ATS-plus-manual workflow that holds up to a hiring conversation, and the moments when a quick TextSight scan is the right cross-check.
Cover letters are the only document in the application packet where voice carries the full load. A flagged resume is recoverable on the interview; a flagged cover letter routes the application to a lower review tier before the conversation happens.
A resume is structured data: roles, dates, accomplishments. Recruiters expect a measured professional register there, and AI-rewritten bullets can read fine if the underlying work is real. The cover letter is the opposite. It exists precisely because the rest of the packet does not have room for voice, motivation, or genuine connection to the role. When the cover letter reads as ChatGPT-generated, the recruiter assumes the candidate could not be bothered to write the one document where their voice was supposed to be the entire point. That assumption is a hiring signal even with no formal AI policy.
Most AI cover letters share four structural features the eye picks up immediately: a 25-word opener stating interest, a middle paragraph of generic capability claims, a closing paragraph asking for an interview, and a uniform three-paragraph length. A trained recruiter spots the shape in roughly three seconds. The first sentence usually settles the read; the closing paragraph confirms it. The rest of the screen is checking specifics before deciding whether to move the application forward.
Through 2025 Workday, Greenhouse, Lever, and iCIMS added AI-content signals to the candidate review layer. The ATS score gives a first-pass filter, and the recruiter eye gives the verdict. The honest ceiling on the manual screen alone is around 85 to 90 percent above 250 words, much higher than email. The ATS flag plus three manual tells gives a defensible read for the hiring conversation; either signal alone is weaker.
Any single tell can appear in a real letter. The signal is when two or three cluster, especially when one lives in the opener and one in the closing. Three or more alongside a TextSight score above 60 percent is a defensible verdict.
An almost unmistakable AI opener. Variants include I am writing to express my interest in, I am thrilled to submit my application, and I am eager to be considered for. The opener fires on the majority of ChatGPT cover letters and almost never appears in a candidate who actually researched the role. Real candidates open with a specific reason for wanting this role at this company, not a templated statement of enthusiasm. One sighting is suggestive; paired with tell 3 in the same paragraph it is a verdict.
Generic value-add language that restates the job posting in slightly different words. Phrases like tailored to your needs, perfectly aligned with your mission, a strong fit for your culture, and well-suited to your requirements appear in ChatGPT drafts because the prompt usually includes the job description and the model loops it back. A candidate who actually read the JD picks one duty and ties it to a past project with a name or a number. Generic alignment language without a specific tie is the body equivalent of the I-am-excited opener.
I am a passionate and dedicated marketing professional with a proven track record. The combination of three or four stacked adjectives (passionate, dedicated, driven, results-oriented, detail-oriented) followed by a generic capability claim is a ChatGPT signature. Real candidates lead with a concrete role title and a specific achievement. The personality summary without a story is the loudest body-paragraph tell after templated alignment language.
The single strongest tell after the opener. Read the entire letter and ask whether the candidate names your product, your last funding round, a team you announced, a customer you signed, or any artefact that proves they spent fifteen minutes on your website. AI cover letters answer no on all counts; they describe being capable in the abstract because the model never visited your site. A letter that could be sent unmodified to any company in your sector is the opposite of a real cover letter, regardless of how the prose was produced.
Opener paragraph of 80 to 110 words, middle paragraph of 90 to 130 words, closing paragraph of 70 to 100 words, all roughly balanced. ChatGPT almost always hits this exact shape. Real candidates vary, sometimes a four-paragraph structure with a separate why-this-company section, sometimes one long anecdote, sometimes a short punchy three-sentence intro followed by a longer body. Uniform balance plus the I-would-welcome-the-opportunity closer is the strongest cluster a recruiter can scan in under five seconds.
Roughly fifty seconds per application once the workflow is built into muscle memory. Steps 1 to 3 are the eye; step 4 is TextSight when the read is unclear; step 5 ties the eye and the tool together for the cases where evidence has to survive a conversation.
Read the first two sentences. AI almost always opens with I am excited to apply for, I am writing to express my interest in, or I am thrilled to submit my application. Real candidates more often open with a specific reason: a product they used, a hire they followed, a problem your team announced. The opener is the loudest single signal and the fastest one to check. A templated opener combined with a missing company reference in the first paragraph is enough to move the letter to the second-pass pile on its own.
Scan the body. Does the candidate name your product, your team, your customers, a recent announcement, or any artefact that proves they read past the job title? AI cover letters describe being a strong fit in the abstract because the model never visited your site. Real candidates land at least one concrete reference, and ChatGPT-assisted candidates who did the research can layer it on top of cleaner prose without the AI tells. Lack of specifics is the body-paragraph equivalent of the templated opener.
Watch the middle paragraph for Furthermore, Moreover, Additionally, In addition, In summary, To conclude. AI uses these to glue templated sentences across the body the way real candidates rarely do. Real candidates trust the paragraph break to do the work and use punchier transitions when they use any at all. Two or more generic transitions in the same paragraph is a strong AI signal, and three or more is a verdict.
Paste the cover letter into TextSight. Free for three scans a day with no signup. Read the overall score, then read the per-sentence highlights. Clustered red sentences in the opener and the closing paragraph confirm the manual read; scattered highlights in a single line are usually a false positive on a borderline sentence. Cover letters of 250 to 400 words sit well inside the calibrated range, and the per-sentence map gives recruiters specific evidence to cite if the hiring conversation requires it.
Go back to the letter and check it against all five cover-letter tells: I-am-excited-to opener, tailored-to-your-needs templating, generic enthusiasm with stacked adjectives, no specific company or role detail, and uniform three-paragraph structure. If three or more tells appear alongside a TextSight score above 60 percent, the verdict is defensible. If the score is high but only one tell appears, the case is weaker; reach for a second independent classifier or a conversation about process before drawing a conclusion.
Free includes 3 detector scans a day. Paid tiers raise the quotas and add the Chrome extension, file upload, and REST API for hiring teams that screen at volume. Yearly billing saves 25%.
Billed $89.88/year — Save $30
Billed $179.88/year — Save $60
Billed $359.88/year — Save $120
Yearly billing saves 25%. View full pricing
No single signal carries a hiring decision on its own. The ATS flag handles volume, the recruiter eye handles judgement, and TextSight handles the borderline cases where evidence has to survive a conversation. Each layer covers the others' blind spots.
Workday, Greenhouse, Lever, and iCIMS all expose an AI-content signal in the candidate review layer through their 2025 integrations. The ATS score is a percentage, not a verdict; it routes high-flagged letters to a second-pass pile without making a final call. The recruiter still owns the read. The right use of the ATS layer is as a sorting hint, not as a rejection switch.
The five tells in this guide are the manual layer. A trained recruiter clears the read in under thirty seconds per letter and flags two or three tells per AI draft on the first scan. The eye also catches the cases the ATS misses: a heavily polished AI draft where the writer swapped the signature vocabulary but kept the rhythm, or an ESL candidate whose uniform structure trips a flag without trace of model authorship in the substance.
For close calls, two finalists tied on resume strength, or any case where the hiring conversation will require evidence, a TextSight scan returns sentence-level highlights and an overall score. The per-sentence map shows exactly which lines came from a model, which converts a gut read into a defensible artefact you can cite. Cover letters of 250 to 400 words sit well inside the calibrated range, and the closing paragraph is usually where the strongest signal cluster lands.
The five tells are calibrated for typical AI cover letters from typical applicants. Several legitimate candidate types trip the same patterns without using AI. Read this section before applying the workflow to any hiring queue with international, junior, or coached candidates.
Most universities teach a three-paragraph cover-letter format that predates ChatGPT and looks similar on the surface. Tells 1, 3, and 5 can fire on a perfectly human letter from a recent graduate using the career-centre template. The specific-anecdote check still works: a real candidate names a class project, an internship deliverable, or a club role with a concrete outcome, even inside a templated skeleton.
A candidate who paid for a coach often gets a letter that hits the same templated shape as an AI draft. Coaches teach the safe structure. The differentiator again is anecdote density. A coach helps shape the prose around real stories; a model invents capabilities in the abstract. If the body contains specific company references, the structural tells are noise.
An ESL candidate who wrote a real letter in their first language and used ChatGPT to translate will trip tells 3 and 5 in the prose layer while the underlying substance is genuinely theirs. The fairness call is the same as for any ESL candidate: weight the structural tells down and lean on tell 4, which measures whether the candidate did the thinking rather than how the prose sounds.
The candidate wrote the substance, then asked Claude or ChatGPT to tighten the prose. The result reads as AI in the sentence-level signal, but the anecdotes are real and the company references are specific. Lean on tell 4, not tells 1, 3, or 5. A letter with real substance under AI polish is closer to a professional edit than to a fabricated application.
The general-purpose sibling guide. Six patterns, a five-step workflow, the ESL caveat, and tool cross-verification.
Read the general guideThe other side of the workflow. Rewrite a ChatGPT cover letter into authentic voice that ATS and recruiters accept.
Read the AI rewriter guideRun the workflow on a real cover letter. Sentence-level highlights, calibrated overall score, bundled Plagiarism Risk.
Open the detectorThe shorter-format companion. Email-specific tells, the three-second screen, and when the signal is too thin.
Read the email guideFree to try, no card. 3 detector scans a day, sentence-level highlights, calibrated for cover letters of 250 to 400 words.